Team Australia in Paris: Five observations from CIGRE 2026

A couple of weeks ago I was fortunate enough to attend my first CIGRE session, held at the Palais des Congrès in Paris from the 23rd to the 28th of August.

CIGRE describes itself as “a global community committed to the collaborative development and sharing of power system expertise”, which in practice means it runs year-round as a network of working groups of professionals from across the world. Most of this collaborative work happens online, but the Paris session is the main event for members to get together in person every two years. Papers are published beforehand, so the discussion starts from a reasonably in-depth starting point, and I particularly enjoyed that the sessions are run efficiently — with a focus on collaborative discussion, rather than just presentations. 

For this belated article, I’ve gone back through my notes to pull out a few of my overarching takeaways from my week — particularly from an Australian point of view.

My own interests (which align with the focus of this blog) meant that I spent most of my week in CIGRE’s C study committees, which cover system development, operations and markets — so this is an admittedly skewed summary. It was great to see global progress on other power system issues, especially from my fellow aussies who seem to be kicking goals when it comes to distribution networks, substations, grid-forming technologies and more — so my apologies that I won’t be covering much of that work in this article.

1. The main transition challenges are shifting from substitution problems to interaction problems

In the opening session on Day 1, Julia Matevosyan, Chief Engineer at ESIG, said the biggest challenge for system planners is no longer how to transition from thermal generation to inverter-based resources (IBRs), but rather how all of those inverter-based resources interact with each other. She went on to elaborate that working out how thousands of separately designed control systems will behave together is a far more wicked (and new) problem than the largely deterministic planning exercises that the electricity industry has spent decades refining.

On that same panel Renuka Chatterjee from MISO (the market operator across the US midwest) talked about the shift from deterministic to probabilistic frameworks for system/market operators, and made the point that the biggest focus for her organisaton’s real-time operations team each day is now aggregate system risk. I also noted down that she went on to say that battery state of charge is emerging as one of the most important control settings that market operators need in order to run the grid reliably under both normal and contingency conditions.

Another shifting energy transition paradigm was highlighted in a poster by Jacqui Bridge, Jonathan Dennis and their colleagues at Queensland’s Powerlink on Day 2 of the session. One of the themes from their poster was that the historical split between generation vs load is slowly fading — particularly as more transmission-connected loads look to install behind-the-meter storage (which in theory would make them behave more like scheduled plant). Their work examined two specific network services those loads could be paid for: (1) curtailment during congestion, and (2) a faster runback following a contingency. That poster included a chart of a particular peak demand event in Queensland in the summer of 2025, showing that network stress and market stress arrived at different times (a theme which I’ll come back to later in this article).

On how to work around these shifting challenges, I noted down a comment from Damir Novosel of Dash Power Partners in the US, speaking at a C5 workshop on Day 3. He said “Five years ago every conference I went to was about IBRs, IBRs, IBRs. Now every conference is data centres, data centres, data centres. In five years it will be something else again”. His argument was that there is little point redesigning the system around whatever the current problem happens to be, and that the better aim should be to make the system robust enough to absorb the next problem.

2. Australia is ahead of the curve, and remains a case study in many ways

Amid the general doom and gloom of the domestic energy news cycle, it’s easy to lose sight of how much progress we’ve made. So it was somewhat comforting to be reminded during CIGRE that there are plenty of areas where Australia is a global leader.

Distribution-level solar is the obvious enduring example, where our rooftop solar penetration is still the highest in the world — and home batteries now appear to be following a similar trajectory. But in my area of interest (the utility-scale stuff), the NEM is ahead of the curve in solving many of the emerging challenges.

As I noted in my summary of COP29 a couple of years ago, among countries decarbonising their electricity system with so little hydro or nuclear in the mix (and no interconnection to neighbours that have it), we are a genuine world leader. As a result of this high IBR penetration and particular interconnection constraints, we face challenges that many other power systems haven’t had to tackle, at least not yet. One example came from Devinda Perera of ElectraNet, presenting on the latest SIPS work in South Australia during a C2 tutorial on Day 1. 

By the working group’s definition, System Integrity Protection Schemes are sets of pre-planned control actions that trigger automatically when a power system hits defined critical conditions (e.g. tripping load or generation, or calling on a battery, in order to stop a local fault cascading into a system black event). During the tutorial, the NEM was compared alongside Norway, Brazil and Spain. Most SIPS are designed to get more out of the existing grid, either by allowing the system to operate above N-1 or by integrating more renewables, whereas South Australia’s latest scheme — with the catchy name of the ‘South Australia Interconnector Trip Remedial Action Scheme’ or SAITRAS — is being developed to prevent uncontrolled islanding.

Once PEC is in service alongside the Heywood interconnector, a non-credible loss of either interconnector under high transfer can cascade into the other one tripping, which would leave SA facing voltage and frequency collapse. Part of the complexity of the scheme is that 15 substations across three states (and hence three different TNSPs) have to be monitored — and those substations live in a transmission corridor that already has several other SIPS operating in it. Devinda stated that the fast component of SAITRAS has to act in under 200 ms, but more importantly, it has to size its response carefully, because over-responding can itself cause a system-wide frequency collapse. Sizing the response is difficult because both interconnectors can have a large import/export limit relative to SA demand, so the amount of load or generation the scheme needs to shed is correspondingly large. As a member of my team pointed out, Tasmania will face somewhat similar challenges once Marinus Link is operating, albeit with two HVDC interconnectors. 

While this SIPS work might be a very specific example, it shows the sort of complex interactions now emerging between operating layers and overlapping protection schemes, which I alluded to in my first theme above.

3. Large disturbances around the world are exposing new risks for system operators

Day 1 included a workshop on large disturbances, which included presentations about a number of recent system black events from around the world, including the Chilean blackout of February 2025 and the Iberian Peninsula blackout in April 2025.

The Chilean case was presented by Rodrigo Espinoza Vilches, an operations manager at Coordinador Eléctrico Nacional, the country’s market operator. The Chilean grid bears some similarities to the NEM in that it is a very long and stringy system, with roughly 40,000 km of transmission lines. In the lead-up to this particular event the grid was operating comfortably inside N-1. Both circuits of a 2 x 500kV north-to-south line were disconnected when the line’s protection system maloperated, without a fault occurring on the network. A communications module associated with that protection had already been reported as disabled prior to the disconnection. Two islands formed on either side of the line, and it took only four seconds for the split to lead to a total loss of supply. Rodrigo put up a slide he described as his most important, comparing a simulation of the central-south island assuming every protection scheme acted correctly and all generation complied with market rules/grid code, against what was actually recorded. The two graphs he showed diverged sharply after the initial incident, and his conclusion was that the island should have remained stable and the fault should never have propagated to a total loss of supply — this highlighted the difference between modelled behaviour and real world outcomes.

A poster in the next day’s C5 session, from Antonia Jorquera Cruz, a lawyer in Chile, was more critical of the role of the various system operators during the event. She noted that the transmission lines had been running above the limits their contingency processes were built around — and that a number of contingency processes were not in place. Her poster went on to critique the investigation process itself — given that the market operator prepares the failure report, she argued that it is hard to treat the investigation as an independent review. And because the operator’s costs are recovered through tariffs, any penalty imposed on it would flow back to the same consumers who experienced the blackout. 

The Iberian case had four presenters, led by Giorgio Giannuzzi of Terna, all of them were members of the expert panel legally responsible for investigating the event. Giorgio described the blackout as the most severe on the European system in over twenty years and the first ever caused by overvoltage. As in the Chilean case, market conditions beforehand were relatively calm. On WattClarity we’ve already collated reports and coverage surrounding the blackout, but as a refresher on the key takeaways, I will quickly relay the high-level sequence of events (as outlined by the four presenters). The first sign of unusual voltage levels came a couple of hours before the blackout, when voltages across Spain’s 400kV network began swinging more than usual. They stayed inside the operational range, but the panel identified the narrow margin between the defined range and the level at which plant disconnects as one of the reasons the event escalated. At around midday, a forced oscillation occurred, followed by a second one. Countermeasures damped both, but left voltage rising — and it was that voltage rise which triggered a cascade of overvoltage disconnections across south-west Spain, taking out 355 MW, then 727 MW, then 928 MW of generation in the space of twenty seconds. Automatic load shedding was activated but, being designed for frequency events, it could not arrest a blackout that was driven by voltage (and the panel’s view was that system inertia was not critical to this particular incident). The root cause tree in the presentation/report listed no single cause, but six contributing factors – among them a low margin between the voltage limit and the disconnection voltage at 400kV, no reactive power response from renewable generators as voltage changed, and ineffective voltage control support from conventional generators, where the requirements do not factor in dynamic behaviour and there is no economic consequence for not meeting these requirements.

From an Australian perspective, it stood out to me that neither event happened during a period of market stress — both were mild days with prices and system conditions appearing relatively typical. It also brought me back to the aggregate system risk highlighted by Renuka, and mentioned above — and to the question of whether N-1 remains sufficient in the face of these more complicated failure modes.

4. No one really knows how to prepare for data centres yet

Data centres came up in almost every session I sat in, as they have in almost every conference I’ve attended this year. Day 4 had a C5 workshop dedicated to data centre impacts on market design and operations. Near the beginning of the workshop, Damir Novosel from the panel asked everyone in the room who works for a data centre company to put their hand up. There were more than 100 people in the room and only five hands went up. He went on to make the point that nearly all of the interest about how the broader power system will handle data centres is being led by people who don’t work for data centre companies. He and Drazena Brocilo, an engineer formerly of Meta and Google, both said they find it difficult to get people from data centre companies onto their panels, usually because these companies are not willing to divulge commercially sensitive information to competitors.

Drazena’s presentation was about the operational characteristics of some of the data centres she has worked on in the past. She highlighted that because AI training runs in bursts, power draw jumps up and down rather than sitting flat — one of her slides showed one data centre load climbing 24 MW in half a second. She said that some of that smooths out with size, because an individual rack might vary by as much as a factor of seven, but when averaged over a few hundred racks across a building those swings are far less extreme. What doesn’t smooth out, she said, is that the pattern changes every time a new AI model is released, and the clusters themselves have grown from a few thousand GPUs to well over a hundred thousand. Manufacturers regularly ask for load profiles of the data centres she is working on, and she tells them she has no idea, and that whatever she does send them will become out of date quickly anyway.

The workshop also went through an example from Dominion in the US, where a transmission line fault was cleared normally by a protection system, exactly as designed. As part of the protection response, the data centres on that part of the network responded by switching about 3.8 GW of load across to their own onsite backup supplies — subsequently, 1.5 GW of that load didn’t come back after the fault passed because the centres’ UPS system was configured differently than expected. As a result, the market operator was left with a large hole in demand.

Anant Venkateswaran summarised these challenges, arguing that we keep treating data centre challenges as separate regulatory, market and engineering issues. He said this is what happened with the smart grid, and that we ended up with a lot of problems we didn’t need to have as a result.

5. No end in sight for the enduring electricity market challenges

Back in 2011 on this blog, Paul wrote about the six enduring challenges for competitive electricity markets, drawing on a list of issues he had helped compile for the Association of Power Exchanges back in 1997. Those challenges appear to have endured for nearly thirty years now — and one of my overarching observations from the week at CIGRE was that the looming impact of data centres may exacerbate many of them.

In a panel session on Day 1, Vinay Sewdien from Krios Infrastructure made the point that around 85% of new large data centres around the world are being built within traditional financial hubs (and of course, those hubs are not usually where the generation is sited). The mismatch between where generation sits and where demand sits ran through most of the C5 posters on Day 2, including a TEPCO paper proposing a distribution-level nodal market across Tokyo’s 1,400 distribution substations.

It occurred to me throughout several workshops and poster sessions during the week, that many capacity markets around the world have a locational or congestion element tied to the capacity payment. In a C5 group discussion meeting on Day 3, Victor Fransisco from PSC explained that even in the WEM, capacity credits are capped by each facility’s Network Access Quantity, which is AEMO’s forecast of the network access actually available to it. It made me think that perhaps the NEM stands out in that our patchwork of informal capacity mechanisms – e.g. the forward market, the CIS, RERT and so on – have no locational element tied to the payment. While the CIS assesses congestion as a merit criterion during awardee selection, there is no ongoing pricing mechanism attached to the underwriting itself (keep in mind CIS contracts can last for 15 years, by which time congestion patterns are likely to have changed).

Transmission investment is another of those enduring challenges, and it took up much of a C1 discussion group meeting I attended on Day 5. A member of the audience from the UK used the Western Isles HVDC link to make the case for building critical transmission projects ahead of modelled need. He noted that the project was first proposed in 2005 and is only now being built — his argument was that the cost of building the project now is likely higher than if the project was instead built too early. In that meeting another audience member from the US spoke about the deficiencies of LCOE as a planning metric, given it takes no account of ancillary services, spinning reserve, storage and so on – a critique that anyone following the ISP debate here will have heard before. 

–

All up it was a very worthwhile week at the session. I am very grateful that the CIGRE community is welcoming and generous with its time, and it was also great to see many of our hard-working NEM colleagues on the world stage.